[Gillian Catriona Ramchand] Verb Meaning and the L(BookZZ.org)

ii

This page intentionally left blank

V E R B M E A N I N G A N D T H E L E X I C O N

The relationship between the meaning of words and the structure of sen-tences is an important area of research in linguistics. Studying the connectionsbetween lexical-conceptual meaning and event-structural relations, this bookarrives at a modular classication of verb types within English and acrosslanguages. Ramchand argues that lexical-encyclopedic content and structuralaspects of meaning need to be systematically distinguished, and that thematicand aspectual relations belong to the latter domain of meaning. The book pro-poses a syntactic decompositional view of core verbal meaning, and sets out toaccount for the variability and systematicity of argument-structure realizationacross verb types. It also proposes a novel view of lexical insertion.

gillian catriona ramchand is Professor of Linguistics at theUniversity of Troms. Her previous publications include Aspect and Predi-cation (1997) and Handbook of Linguistic Interfaces (2006, with CharlesReiss).

CAMBRIDGE STUDIES IN LINGUISTICS

General Editors: p. austin, j. bresnan, b. comrie,s. crain, w. dressler, c. j. ewen, r. lass,d. lightfoot, k. rice, i. roberts, s. romaine,n. v. smith

Verb Meaning and the LexiconA First-Phase Syntax

In this series

71 knud lambrecht: Information structure and sentence form: topic, focus, and the mental representa-tion of discourse referents

72 luigi burzio: Principles of English stress73 john a. hawkins: A performance theory of order and constituency74 alice c. harris and lyle campbell: Historical syntax in cross-linguistic perspective75 liliane haegeman: The syntax of negation76 paul gorrel: Syntax and parsing77 guglielmo cinque: Italian syntax and universal grammar78 henry smith: Restrictiveness in case theory79 d. robert ladd: Intonational morphology80 andrea moro: The raising of predicates: predicative noun phrases and the theory of clause structure81 roger lass: Historical linguistics and language change82 john m. anderson: A notional theory of syntactic categories83 bernd heine: Possession: cognitive sources, forces and grammaticalization84 nomi erteschik-shir: The dynamics of focus structure85 john coleman: Phonological representations: their names, forms and powers86 christina y. bethin: Slavic prosody: language change and phonological theory87 barbara dancygier: Conditionals and prediction88 claire lefebvre: Creole genesis and the acquisition of grammar: the case of Haitian creole89 heinz giegerich: Lexical strata in English90 keren rice: Morpheme order and semantic scope91 april mcmahon: Lexical phonology and the history of English92 matthew y. chen: Tone Sandhi: patterns across Chinese dialects93 gregory t. stump: Inectional morphology: a theory of paradigm structure94 joan bybee: Phonology and language use95 laurie bauer: Morphological productivity96 thomas ernst: The syntax of adjuncts97 elizabeth closs traugott and richard b. dasher: Regularity in semantic change98 maya hickmann: Childrens discourse: person, space and time across languages99 diane blakemore: Relevance and linguistic meaning: the semantics and pragmatics of discourse

markers100 ian roberts and anna roussou: Syntactic change: a minimalist approach to grammaticalization101 donka minkova: Alliteration and sound change in early English102 mark c. baker: Lexical categories: verbs, nouns and adjectives103 carlota s. smith: Modes of discourse: the local structure of texts104 rochelle lieber: Morphology and lexical semantics105 holger diessel: The acquisition of complex sentences106 sharon inkelas and cheryl zoll: Reduplication: doubling in morphology107 susan edwards: Fluent aphasia108 barbara dancygier and eve sweetser: Mental spaces in grammar: conditional constructions109 hew baerman, dunstan brown and greville g. corbett: The syntaxmorphology inter-

face: a study of syncretism110 marcus tomalin: Linguistics and the formal sciences: the origins of generative grammar111 samuel d. epstein and t. daniel seely: Derivations in minimalism112 paul de lacy: Markedness: reduction and preservation in phonology113 yehuda n. falk: Subjects and their properties114 p. h. matthews: Syntactic relations: a critical survey115 mark c. baker: The syntax of agreement and concord116 gillian catriona ramchand: Verb meaning and the lexicon: a rst-phase syntax

Earlier issues not listed are also available

V E R B M E A N I N G A N DT H E L E X I C O NA FIRST-PHASE SYNTAX

G I L L I A N C AT R I O NA R A M C H A N DUniversity of Troms, CASTL

CAMBRIDGE UNIVERSITY PRESS

Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, So Paulo

Cambridge University PressThe Edinburgh Building, Cambridge CB2 8RU, UK

First published in print format

ISBN-13 978-0-521-84240-2

ISBN-13 978-0-511-38791-3

Gillian Catriona Ramchand 2008

2008

Information on this title: www.cambridge.org/9780521842402

This publication is in copyright. Subject to statutory exception and to the provision of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press.

Cambridge University Press has no responsibility for the persistence or accuracy of urls for external or third-party internet websites referred to in this publication, and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

Published in the United States of America by Cambridge University Press, New York

www.cambridge.org

eBook (NetLibrary)

hardback

Contents

Acknowledgements page ixAbbreviations x

1 Introduction 11.1 Capturing argument-structure generalizations 5

1.1.1 The lexicalthematic approach 51.1.2 Generativeconstructivist approaches 9

1.2 Excursus on the role of encyclopedic knowledge 12

2 The empirical ground 182.1 Selection versus variability 18

2.1.1 Causation 232.1.2 Telicity 252.1.3 Nonaspectual arguments 33

3 A rst-phase syntax 383.1 The semantic interpretation of structure 42

3.1.1 Rhematic material 463.1.2 Agents and experiencers: the special case of mental states 533.1.3 Stative predications 55

3.2 Integrating the verbal lexical item 57

4 Deriving verb classes 634.1 Initiationprocess verbs 63

4.1.1 Transitives 644.1.2 Intransitives 71

4.2 Initiationprocessresult verbs 744.2.1 Transitives 744.2.2 Intransitives 78

4.3 Transitivity alternations 824.3.1 Degree achievements 89

4.4 Conation verbs 91

vii

viii Contents

4.5 Double object verbs 1004.5.1 Applicatives more generally 105

4.6 Statives 1064.7 Summary 107

5 Paths and results 1105.1 PPs: paths and places 1105.2 APs: paths and results 1215.3 Types of resultatives in the rst phase 1255.4 The verbparticle construction 1315.5 Russian lexical prexes 1385.6 Completive complex predicates in Indic 1425.7 Conclusion 149

6 Causativization 1506.1 Introduction 1506.2 An analytic causative in Hindi/Urdu 1516.3 Overview of Hindi/Urdu transitivity alternations 154

6.3.1 Transitiveintransitive pairs via vowel alternation 1556.3.2 Causativization using the -aa sufx 1566.3.3 Causativization using the -vaa sufx 1616.3.4 Status of the causee 1656.3.5 Status of causer 1666.3.6 Summary 167

6.4 Analysis 1686.4.1 Representing the verb classes in Hindi/Urdu 1686.4.2 Direct vs. indirect causation 1696.4.3 Direct causativization in -aa 1716.4.4 Indirect causativization in -vaa 1766.4.5 Event underassociation and the intermediate agent/causee 1816.4.6 Consequences 187

6.5 Reinterpreting internal and external causation 1886.6 Conclusion 192

7 Conclusion 1937.1 Summary of the system 1937.2 The connection to tense 1987.3 Open questions 203

References 205Index 215

Acknowledgements

This book could not have been written without the generous engagement ofmany of my friends and colleagues in the eld. It would be impossible toname them all, or to do justice to all their comments and criticisms. Forlengthy discussions, arguments and advice I would like to single out DavidAdger, Miriam Butt, Raffaella Folli and Peter Svenonius. I also thank CASTL(Center for the Advanced Study of Theoretical Linguistics) and the Universityof Troms for being one of the best and most stimulating places in which towrite a book. Much of the material for this book arose out of interaction withcolleagues and students during the course of seminars and lectures. My debt toall of those interactions and the other research going on at Troms can be seenclearly in the pages of this book. For the nal stages of writing and revising,I am once again grateful to Peter Svenonius, who commented on more draftsthan time or sanity should allow.

ix

Abbreviations

The following abbreviations are used in the glosses:

ACC accusative caseCAUSE causativeCLASS classierDAT dative caseDEC declarativeDIR directional markerERG ergative caseF feminine agreementFUT future tenseIMP imperfectiveINF innitival formINSTR instrumental caseLOC locative markerM masculine agreementNOM nominative caseOBL oblique casePASS passivePAST past tensePERF perfectivePERFPART perfect participial formPL plural agreementPRES present tensePROG progressiveSG singular agreementTOP topic1 rst-person agreement3 third-person agreement

x

1 Introduction

Classical generative grammar partitions linguistic competence into three basiccomponents: lexical knowledge, phrase structure rules and transformationalrules (Chomsky 1965, 1981). One of the fundamental debates over the years,and one which is still alive today, concerns the division of labour between infor-mation and processes that reside in the lexicon and those rules and processesthat are part of syntax.

In this book, I explore a view of the architecture of grammar whereby thelexicon is eliminated as a module with its own special primitives and modesof combination. By this, I do not intend to deny that there are items withinthe language that need to be listed/memorized, or that they are associated withgrammatical information. Rather, I will seek to claim that to the extent thatlexical behaviour is systematic and generalizable, this is due to syntactic modesof combination and not to distinct lexicon-internal processes (Hale and Keyser1993, etc.). The general ideology is not novel; I am attempting to implementan old idea in the light of current, accumulated knowledge concerning thenature of lexicalgeneralizations and patterns. In pursuing, as I will, a radicallyunstructured viewof the lexicon, I engagewith recent ideas of constructionalism(Goldberg 1995,Marantz 1997b, Borer 2005) andmakemy own proposal basedon what I take to be the core empirical issues of thematic roles, event structure(aktionsart) and selection.

One of the things I will take for granted in this work is that human beings lin-guistic competence includes, minimally and crucially, a (linguistically specic)combinatorial system.1 It is this combinatorial system that I will be referring towith the term syntax, and I will assume that the system itself is universal, in

1 Here I also wish to abstract away from the debate concerningwhether this combinatorialsystem is representationally innate in the sense of all the basic knowledge existing in ahardwired repository of brain structure, or whether it emerges inexorably as a result ofthe learning strategies abstractly encoded in a language acquisition device. In fact, it isnot even relevant to my argumentation whether the combinatorial system that emergesis specic to language, or whether it is part of a more general human symbolic capacity.

1

2 Introduction

the sense of underlying all instantiations of human language. Under the viewI will be pursuing here (and one that is implicit in much work within minimalistsyntax, and even earlier), this is the only linguistically relevant combinatorialsystem that there is, i.e. we are dealing with only one set of primitives and oneset of operations.

Two distinct types of lexical information have always been recognized:unstructured encyclopedic information with its innitely variable web of asso-ciation and nuance; and the grammatically relevant, more systematic, class ofinformation that interfaces with the syntactic system (Chomsky 1965, Jackend-off 1983). The classical assumption has been that two such types of meaningcoexist in a module that is termed the lexicon, with the latter level being thelinguistically relevant subset of the former (cf. Levin and Rappaport Hovav1995, Levin and Rappaport 1998).

Taking the existence of a lexical module of some sort for granted, manyearly generativist debates were concerned with the location within the gram-mar of particular sorts of linguistic generalizations, i.e. whether they shouldmore properly be considered lexical or syntactic (see Chomsky 1970 forfoundational early discussion, as well as later debates staged in Baker 1988and Belletti and Rizzi 1988 vs. Alsina 1992 and Bresnan and Moshi 1990).Importantly, claiming that there are generalizations that can only be stated atthe level of lexical information is different from merely accepting that lexicalitems possess syntactic information, hence the debate. In general, some theoriessuch as LFG sought to establish the validity of separate modules with their ownprimitives and modes of combination, linked by correspondence rule (Bresnan1982); whereas GB theory and its descendants took the view that the lexiconshould be seen as the repository of essentially idiosyncratic/memorized infor-mation with no independent combinatorial primitives (Di Sciullo andWilliams1987, Chomsky 1981). It is a version of the latter position that I will be arguingfor in this book, although the details prove stickier than one might imagine ifone is intent on not begging the important questions.

The main challenge to the unstructured lexicon view has always been theexistence of thematic, or argument-structure generalizations,2 captured in GBtheory via the D-structure level of representation, or by Hale and Keyser viaL-syntax (an encapsulated syntax for the building of lexical items). In morerecent minimalist work (Chomsky 1995, 2000), presumably no such additional

2 I am concerningmyself purely with syntax here. Lexical phonology, if it exists as a set ofoperations distinct from postlexical phonology,might constitute another such challenge.I will assume optimistically, for the purposes of this book, that those challenges canalso be overcome.

Introduction 3

level of representation can exist, but the operation of initialMerge is potentiallyavailable as a locus for these generalizations. Since this operation is triggeredby selectional features (Chomsky 1995), capturing generalizations at this levelwill depend on the nature of the features involved, and the nature of selectionand insertion of lexical items.

The key here is therefore the features on lexical items and how they might bedeployed to create selectional generalizations.One approach to the problem is todeny that such selectional generalizations exist. This is the view most recentlytaken by Marantz (1997b) and (1998, 2005), whereby lexical items possessno syntactically relevant information that could constitute a constraint on theirinsertion possibilities (not even category information). The actual limits on vari-ability reported in more standard accounts would then have to be due to limitsbased on real-world knowledge and convention (extralinguistic).While I will besympathetic to the attempt to void the lexicon of argument-structure informationand processes, I will still seek to encode some notion of selectional informationthat constrains the way lexical items can be associated with syntactic structure(so in this sense I will consider myself responsible for at least some of the datacited by the lexicalist camp, e.g. Levin and Rappaport 1998, Reinhart 2002).

In order to frame the particular proposals of this book more concretely, it isuseful to compare schematic versions of the architecture of the grammar withrespect to the lexicon that have emerged either explicitly or implicitly over theyears. My descriptions of the main options are not necessarily specic to aparticular researcher, although I will attempt to associate the different abstractpositions with various prominent proposals in the literature. Every individualproposal has its own subtleties and makes specic decisions about implementa-tion, which I will abstract away from here. The purpose in what follows, rather,is to characterize the extreme options in an idealized way, in order to clarifywhat is at stake, and to contextualize the view I will develop in this book.

The core questions that any theory of the lexicon must address are thefollowing:

(i) Is the lexicon a module of the grammatical system, with its owndesignated primitives and operations?

(ii) If the answer to (i) is yes, what is the division of labour betweenlexicaloperations and the recursive/generative syntactic computation(which must exist, by hypothesis)? 3

3 See, for example, Wasow (1977) for an argument for the lexicon-internal treatment ofpassive, and Dubinsky and Simango (1996) for a discussion of adjectival passives inEnglish and Chichewa, also Marantz (2001) for a recent reassessment.

4 Introduction

(iii) What is the relationship between lexical information and nonlanguagededicated parts of the mind/brain?

According to a common-sense standard view of lexical entries, lexical itemsused in language contain both language-specic and non-language-specicmemorized information. I represent a possible listing in the toy lexical entryin (1).

(1) run/ r n /

Verb, < 1 >+dynamic; telic

argument 1: Theme; argument 1: animatecontinuous directed motion undergone by < 1 >

motion involves rapid movement of legs,no continuous contact with ground

...

Associations: exercise, boredom, heart attacks

In principle, anything can be memorized; nevertheless, certain lexical entriesdo not exist in natural language. For example, lexical entries where the agentiveinstigator of an action is realized as the direct object,while the passive undergoercomes out as the subject, do not seem to be attested. This kind of pattern isclearly not arbitrary. The generalizations about thematic linking to grammaticalfunction, and the fact that intransitive verbs with more agent-like argumentsbehave linguistically differently from intransitive verbs with more patient-likearguments (the unaccusative hypothesis, Perlmutter 1978), are generalizationswe would like our theory of grammar to capture. There are two clear strategiesfor implementing the generalizations we need:

(I) The lexicalthematic approach, which allows for the semantic classi-cation of role types within the lexicon, readable by a linking theorythat places these different roles in different places within the structure.In this approach, the relevant information is projected from the lexicon.Under this view, the lexicon is a submodule of the language facultysince it has its own distinct primitives and modes of combination.

(II) The generativeconstructivist approach which allows free buildingof syntactic terminals, but allows general encyclopedic knowledge tomediate whether a particular lexical item may be inserted in those

1.1 Capturing argument-structure generalizations 5

terminals or not (Borer 2005, Marantz 2001). Under this view, thelexicon is not a submodule, since it contains no grammatically relevantinformation or processes.

1.1 Capturing argument-structure generalizations

1.1.1 The lexicalthematic approachIf we embark on the rst strategy, and take the lexicon to be a genuine moduledealing with argument structure, then the linguistically relevant part of thelexical entry looks perhaps as follows (withmore or less internal structuring) (2).

(2) run; V

Theme

However, the most important challenge when pursuing this view lies in statingthe correspondence or linking rules between the lexical module and its internalstructuring and the syntactic module and its internal structuring. One traditionalway of doing this includes postulating the existence of a thematic hierarchywhichmediates the assignment of thematic participants to grammatical functionor structural position. Some examples of thematic hierarchies are shown in (3)and (4) below, with examples of rules of argument realization in (5) taken fromLarson (1988).

(3) Larson (1988)Agent < Theme < Goal < Obliques(manner, location, time)

(4) Grimshaw (1990)Agent < Experiencer < Goal/Source Location < Theme

(5) Principle of Argument Realization 1 (Larson 1988)If is a predicate and is an argument of , then must be realizedwithin a projection headed by .Principle of Argument Realization 2 (Larson 1988)If a verb determines -roles 1, 2 n, then the lowest role on theThematic Hierarchy is assigned to the lowest argument in constituentstructure, the next lowest role to the next lowest argument, and so on.

It is important to note that there has not been consensus on the number andtypes of thematic relations the theory should employ, nor on the exact nature

6 Introduction

of the thematic hierarchy involved. Dismay at the lack of reliable and objectivelinguistic diagnostics led at least one researcher, Dowty (1989), to despair ofthe enterprise altogether. Dowty himself offered a more exible alternativeto thematic generalizations in his 1990 article, advocating a more uid kind oflinking based on the relativeweighting of a number of different proto-properties.These are listed in (6) below.

(6) Dowtys proto-roles (1990)Contributing properties for the Agent Proto-role(a) volition(b) sentience (and/or perception)(c) causes event(d) movement(e) referent exists independent of action of verbContributing properties for the Patient Proto-role(a) change of state (including coming into being, going out of being)(b) incremental theme (i.e. determinant of aspect)(c) causally affected by event(d) stationary (relative to movement of Proto-agent)(e) Referent may not exist independent of action of verb, or may not

exist at all.

Dowtys argument selection principle (1990)The argument of a predicate having the greatest number of Proto-agent prop-erties entailed by the meaning of the predicate will, all else being equal, belexicalized as the subject of the predicate; the argument having the greatestnumber of Proto-patient properties will, all else being equal, be lexicalized asthe direct object of the predicate.

In fact, this is even more of a retreat than it appears to be, since the prin-ciple of argument selection given above cannot be seen as a fact about thesynchronic computational system (since, plausibly, decisions about what getsto be the subject are not computed on-line or subject to variability in casesof ties), nor as a fact about memory (if one assumes that memory does notcalculate, but merely retrieves information). Dowtys principle basically givesup the idea that the generalizations we see should be represented in the coregrammar the properties he gives must have the status of general cognitivetendencies which ultimately underlie how various concepts tend to get lexi-calized (memorized) in natural language. Dowtys proto-roles are neverthelessinteresting and instructive, because they are the ones that he judged to be most


criterial of linguistic behaviour. As we will see, I will argue that these gen-eral properties (as opposed to thematic role labels) are in fact the right level ofabstractness for stating systematicities concerning the mapping between syntaxand semantics.

One further view on thematic linking is worth mentioning here, that of Baker(1988) and subsequent work. In Bakers view, thematic roles are linked tostructure/grammatical function not via a relative ranking system as in theo-ries employing the thematic hierarchy, but in a more absolute sense. In otherwords, each type of thematic role has its own special structural position that itis associated with.

(7) The Uniformity of Theta Assignment Hypothesis (UTAH)Identical thematic relationships between items are represented byidentical structural relationships between those items at the level ofD-structure. (from Baker 1988: 46)

In recent work, Baker (1997) claims that the notion of thematic role that isrelevant for this principle is somewhat more abstract than the traditional list,but rather contains such thematic categories as: Agent (specier of the higherVP of a Larsonian structure), theme (specier of the lower VP of a Larsonianstructure), Goal/Path (complement of the lowerVP). Still, the principle (and, infact, many systematic principles of linking) receives its major challenges fromdata pairs such as (8), (9) and (10) below, where apparently identical thematiccongurations are differently aligned in the syntax.

(8) Experiencer object vs. experiencer subject(a) Wolves frighten John.(b) John fears wolves.

(9) The dative/double object alternation(a) John gave the book to Mary.(b) John gave Mary the book.

(10) The spray-load alternation(a) Bill loaded the cart with hay.(b) Bill loaded hay on the cart.

A lexical theory containing linking principles such as those described aboveessentially has three main options in dealing with such exibility. The rstoption is to make the linking principles themselves exible and nondetermin-istic. This is in a sense the option taken by Dowty (1990) and certain versions

8 Introduction

of LFG (cf. Bresnan 2001). The second option is to claim that the (a) and (b)sentences above involve the same underlying congurations, but at least one ofthem involves a nontrivial syntactic derivation. This, for example, is the optiontaken by Larson (1988) in his treatment of the double object alternation, andthe solution advocated by Baker (1997) for one set of alternations as well. Theextent to which this general strategy is plausible will depend on the syntacticprinciples at stake being independently justiable, and not ad hoc additions tothe syntactic tool box merely to save the UTAH and its kin. The third strategy,of course, is to claim that the thematic roles in the (b) sentences are actuallydifferent from those in the (a) sentences (cf. Oehrle 1976, Pesetsky 1995 forthe double object construction). This is in fact the claim Baker (1997) makesfor the spray-load alternation, although not for the double object alternation.The success of this strategy revolves around resolving the tension between theneed to use fairly abstract thematic labels to capture the natural classes whichexist but which are nevertheless subtle enough to distinguish between thematicrelationships in the closely related pairs above.

Thus, assuming a lexicon which contains at least some annotations froma syntactic vocabulary encompasses a wide range of theories from differentideologies, I think it is possible to distinguish two clear extremes.

(i) The static lexiconThe lexicon contains argument-structure information which correlatesin a systematic and possibly deterministicwaywith syntactic structure.The lexicon has its own vocabulary, but there are no lexicon-internalmanipulations prior to insertion. Syntactic transformations can alterthemanifestation of a particular set of lexical information in a sentence.

(ii) The dynamic lexiconThe lexicon contains argument-structure information which correlatesin a systematic and possibly deterministic way with syntactic struc-ture. The lexicon has its own vocabulary, as well as lexicon-internalmanipulations prior to insertion. Syntactic transformations to accountfor alternations are kept to a minimum.

Both types of approach necessitate a linking theory because each moduleuses a different vocabulary, but independent differences also arise relating towhether that linking is assumed to be deterministic and absolute, determin-istic and relative, or even one which involves optionality (nondeterministic).I take Baker (1988) to be a representative of the (deterministic) static lexicon


view, with Levin and Rappaport Hovav (1995) being proponents of the dynamiclexicon view.

Flexibility in verbal meaning exists on the level of aspectual specication aswell, prompting the postulation of lexicon-internal processes such as templateaugmentation (cf. Levin and Rappaport Hovav 1995) and event-type shifting(van Hout 2000a, 2000b).

Thus, while there are many differences of approach within this broad classof theories, the very notion of linking theory presupposes that two distinctvocabularies from two distinct modules are being connected. Which rulesand transformations exist in one or the other, or indeed both modules (thelexicon and the syntax), constitutes an important debate in the context of thiskind of architecture, and has a direct impact on the nature of the labels andnatural classes proposed for the thematic roles as listed in the lexicon. In thisbook, I will pursue the view that there is only one module where rules andtransformations can be stated (I will call this the narrow synsem computation).However, the patterns uncovered through these classic debates will form muchof the descriptive base for the proposal, and the general intuition behind theUTAH, which correlates structure with meaning fairly directly, will be presentin the implementation.The bottom line is that lexical theoriesmust either invokelexicon-internal processes, or tolerate massive stipulated homonymies. To theextent that the processes that need to be assumed can actually be elegantlycaptured in the syntax, it should be preferable on grounds of parsimony toassume only one such system if we can get away with it.

1.1.2 Generativeconstructivist approachesUnder an extreme constructivist view, lexical roots contain no syntacticallyrelevant information at all; they are just bundles of cognitive and encyclopedicinformation. Consider the revised lexical entry below in (11).(11) run

continuous directed motion undergone by animate entitymotion involves rapid movement of legs,

no continuous contact with ground...

Associations: exercise, boredom, heart attacks etc.

The complete lack of syntactic or argument-structure information on the lecard makes it in principle compatible with many different syntactic frames.

10 Introduction

Thus, Borer (2005) offers the following range of examples for the English verbsiren (which signicantly is also compatible with nominal syntactic structure).(12) (a) The re stations sirened throughout the raid.

(b) The factory sirened midday and everyone stopped for lunch.(c) The police sirened the Porsche to a stop.(d) The police car sirened up to the accident.(e) The police car sirened the daylights out of me. (from Borer 2005)

The well-known problem with this view is of course the fact that argument-structure exibility is not as general as it would suggest. For example,some intransitive verbs resist causativization (13a), and others resist telicaugmentation (13b):(13) (a) *John slept the baby.

(b) *John watched Mary bored/to boredom.How does one account for this kind of selective behaviour in a theory

where the lexical item contains nothing written in the syntactic vocabulary?For Borer (2005) the (only internally consistent) answer is given: convention,habits of speech and real-world knowledge make certain combinations of rootplus syntactic/functional information unusable or infelicitous.

Under theBorerian andMarantzian views, the distinction between lexical andfunctional categories hardens, lining up with real-world vs. linguistic meaningrespectively. The root is the only lexical category under these views, althoughironically, of course, it does not even carry category information. All categoryinformation and linguistically manipulable meaning come from the functionalstructure that sits on top of the root. Once again, there are many versions of thisposition out there in the literature, with slightly different choices of functionalprojections and labels for any particular effect. In Borers structure, there is anaspectual quantity phrase that sits on top of the VP and is responsible for bothtelicity and object quantity effects. In Traviss work, there is an event phrase(EP) higher than VP and an aspectual phrase (AspP) sandwiched in betweenLarsonianVP shells, the latter of which is correlated with telicity (Travis 2000).In Ritter and Rosen (1998), there is an initiational aspectual projection on top ofTP, and a delimitational aspectual projection in betweenTP andVP. The generalapproach also varieswith respect to howmuch information is allowed to the lexi-cal root and howmuch is relegated to the functional structure. InKratzer (1996),the lexical root contains information about the internal argument, but the exter-nal argument is introduced by a hierarchically superior functional head v. Theidea of little v in its turn has had many proponents, different types of external


argument being introduced by different avours of the little v head (cf. Harley1995, Folli and Harley 2004). Within this spread of opinion, in a sense thesame debate is being staged the division of labour between the syntax and thelexicon. Once again, we can distinguish two extremes.

(i) The naked roots viewThe root contains no syntactically relevant information, not evencategory features.

(ii) The well-dressed roots viewThe root may contain some syntactic information, ranging from cat-egory information to syntactic selectional information and degrees ofargument-structure information, depending on the particular theory.This information is mapped in a systematic way onto the syntacticrepresentation which directly encodes it.

The latter position is virtually indistinguishable in practice from the staticlexicon view in the section above, and could be made perfectly compati-ble with it provided the technical issue of selection and selectional featuresis decided. In practice, the majority of researchers in the decompositionalor constructivist camp actually fall between the two extremes describedabove. In essence, the theoretical questions revolve around deciding howmuch functional structure (which heads precisely, and in which order(s))related to so-called argument-structure generalizations can be justied inthe syntactic representation. It should be clear that this question correspondsempirically to the question of how many and what type of thematic roles wehave and how they line up with syntactic position in the deterministic staticlexicon view.

In discussing the general class of constructivist approaches, it is necessaryto say a word about Construction Grammar, which I think must be clearlydistinguished from those above, even though it shares with them the view thatstructures carry meaning. The theory of construction grammar, as found forexample in Goldberg (1995), allows that structures carry meaning but seemsto make the opposite architectural claim to that of the constructivists discussedabove: it analogizes constructions to lexical items that have to be listed andmemorized. In this sense, it downplays the generative character of the naturallanguage system and allows large templatic chunks to be simply memorized.The view taken in this bookwill be that the reason constructions havemeaning isbecause they are systematically constructed as part of a generative system (syn-tactic form) that has predictable meaning correlates. Thus, the view proposedhere will be generativeconstructivist in spirit, but not constructionist.

12 Introduction

1.2 Excursus on the role of encyclopedic knowledge

One of the important ideas in this system, which I share with many others, is theneed to make a strict and principled distinction between linguistic meaning andencyclopedic content. In this context, it is useful to examine another prominentview of the lexicon, as espoused by Pustejovsky (1991). Pustejovskys gen-erative lexicon is interesting because it explicitly contains two different sortsof information, which he acknowledges to be different. Specically, Puste-jovsky acknowledges that (i) there is no way that meaning can be divorcedfrom the structure that carries it and (ii) that the meanings of words are alsothe reections of deeper conceptual structures, i.e. the image of nonlinguisticconceptual organizational principles.

Information in the lexicon according to PustejovskyA Argument structure: the behaviour of a word as a function, with its

arity specied. This is the predicate argument structure for a word,which indicates how it maps to syntactic expressions.

B Event structure: identication of the particular event type (in the senseof Vendler) for a word or phrase, e.g. as state, process or transition.

C Qualia structure: the essential attributes of an object as dened bythe lexical item.

The relation between it and its constituent parts constitutive role That which distinguishes it within a larger domain formal role Its purpose and function telic role Whatever brings it about agentive roleD Inheritance structure: how the word is globally related to other

concepts in the lexicon.

Rather than assuming a xed set of primitives in lexical-semantic representa-tions, Pustejovsky assumes a set of generative devices to construct semanticexpressions for one aspect of the lexical representation. These generativedevices are basically located in the event-structure module of Pustejovskyssystem and the recursive rules that combine them.

While I am completely in sympathywith the distinctionmade by Pustejovsky,I differ with respect to the architectural decision that he makes. Basically, sinceevent-structure composition is productive and does not actually need to bemem-orized, it is not clear whether it really belongs in a designated module separatefrom syntactic generative devices proper, i.e. it looks as if these principles needto apply to constructions. If the combinatoric devices proposed are essentially


redundant with syntax, then they do not belong here. During the course of thisbook I will attempt to show that it is a mistake to take argument structure/eventstructure facts as a property of the lexicon, or even single lexical items, sincethe same structural organization can be detected in languages that use singlewords or analytic constructions to express verbal meaning.

The second problem with Pustejovskys position is in a sense the converse ofthe rst. This concerns the amount of cognitive information that is claimed to bespecied in an items lexical entry. I think it can be shown that this is a slipperyslope, and that the effects of qualia structure are not in fact distinguishable inany reliable way from real-world knowledge whose effects are unpredictable.Consider the following example.

(14) (a) John began a book.(began writing it, or began reading it)

The qualia structure of the item book contains information about its telic roleand its agentive role in the above sense, and this is what is supposed to licensethe two different types of inference in (14) above. But, of course, other readingsare possible. If John and Mary are systematically going through all of Billsmagazines erasing all the es, one can say (14a) to indicate the start of thatprocess applied to a book. Given this kind of interpretational possibility, it isnot obviouswhere one stops annotating a lexical itemwith how it can potentiallyinteract with the real world.

In some cases, the effects feeding off qualia structure seem to be evenmore than just specicities of interpretation, but actually have syntactic conse-quences. In the pairs of examples in (15) and (16) below, the only thing changedis the choice of DP object.(15) (a) John baked a cake.

(b) John baked a potato.(16) (a) John painted a picture.

(b) John painted a wall.

Pustejovsky uses the contrast in inferences between bake a cake and bakea potato to argue for different qualia structure for cake and potato intheir lexical entries. On the other hand, real-world knowledge makes equallygreat differences to the inferences licensed. In the case of a potter makingclay miniatures of edible items, the judgement concerning (15b) is ratherdifferent.

At the same time, the creation sense allowed for the bake a cakeand paint apictureexamples is probably in fact a different structure within the verb phrase,

14 Introduction

with further distinct linguistic consequences. So, while the creation sense useof paint can give rise to a benefactive construction, as in (17), or a resultativeconstruction, as in (18), the incremental theme interpretation of an object doesnot, as the (b) examples show.(17) (a) John painted me a picture.

(b) ??John painted me a wall.(18) (a) John painted a wall red.

(b) ??John painted a picture red.

The point here is that there is no a priori way of deciding what goes intothe lexical entry with respect to qualia structure. There is no evidence thatdifferences in inference properties at this level are linguistic at all. Rather,it seems more as if language allows different structures, but the real worlddetermines felicity and detailed inferential patterns.

In this book, I will indeed be taking seriously the distinction between lexical-encyclopedic content and structural correlates of meaning. The decisions aboutwhat kind of meaning fall on which side of this divide is of course a subtle andempirical question, and should not be prejudged. The theoretical point, though,is that if all so-called lexical content can be reduced to either one or the other,the structural-generative aspect of meaning can be protably analysed as part ofthe syntactic component. The lexical-encyclopedic side is a matter for generalcognition. The lexical entry itself is the memorized link between chunks ofconceptual structure and conditions of insertion; it does not need to reside in amodule with its own combinatorial primitives.

The distinction I am making has always been acknowledged in the lexicalistproposals of Levin and Rappaport Hovav (1995), for example. The abstracttemplatic aspects of Levin and Rappaports representations are the ones thatI will give syntactic representation. The insertion of lexical items into thesestructures will be analogous to the association of constants to the variablesof those abstract templates in the lexicalist views. Thus, the lexical item willcontribute conceptual content to structural aspects of meaning, and will betagged with category labels as a way of constraining that insertion. Thus, unlikethe radical generativeconstructivist position, I will not be assuming that lexicalitems are free of syntactic information, and neither will I be assuming that theyare inserted always at the bottom or root of the tree. The pure naked rootsview seems too strong, and only appears to work when it ignores the substantialempirical and technical issues surrounding selection. Moreover, it denes awaythe central property of the lexical item as an associative web of properties fromdifferent modules, including, crucially, the narrow synsem computation.


The entry for run in (19) is an idealization of the distributed nature of theinformation involved. It can be represented in one box, but only as a convenientidealization, because of strong links of mental association.

(19) runLabel seen by PF: / r n /

Label seen by narrow synsem computation: v, Vcontinuous directed motion undergone by animate

motion involves rapid movement of legs,no continuous contact with ground

...

Associations: exercise, boredom, heart attacks

Thus, lexical items in this system will associate to syntactic representationsvia their syntactic labels. Constructional-semantic and lexical-encyclopediccontributions are unied to form a proposition at the interface with thecognitive/interpretive systems of the mind/brain.

Under the kind of view explored in this book, the lexicon cannot exist asa module because it is not encapsulated, but associates representations fromradically different cognitive modules (conceptual, articulatory, formal).

To the extent that variability of use within syntactic structures is systematic,the primitives and processes involved are the same ones that are used by syntax.I will take as a starting point that it is more parsimonious to assume that they arepart of the same system with syntax. On the other hand, there may well be cog-nitive generalizations about conceptual structure, but we know that conceptualstructure must exist outside language. Lexical-encyclopedic knowledge is of apiece with real-world knowledge and does not give systematic compositionaleffects (the crucial distinguishing property of language).

Syntactic category information appears to be unavoidable for mediating theassociation of functionallexical items and syntactic structure. If we can reduceall the selectional constraints of so-called lexical categories to this type ofassociation too, then there is no argument against it from parsimony. As withall proposals concerning the architecture of the system, one makes a choiceas to where the complexity of that system resides. Under this view, there isonly one combinatorial system, and the primitive modes of combination willbe minimalist (i.e. conned to (Re)Merge and Agree, triggered by the need tocheck uninterpretable features), but the complexity will reside in the extendedfunctional sequence assumed in the syntax and the larger set of category featuresthat implies.

16 Introduction

The other main point that I want to argue for in this book concerns thenature of the syntaxsemantics interface. The basic combinatoric system ofthe lowest part of the clause emerges as something which encodes semanticinformation as well as the traditionally syntactic. An inevitable consequence ofthe separation of lexical-encyclopedic information from the structural is thatthe structures themselves will be seen to determine abstract predicational andevent-compositional semantics. However, unlike the constructional grammarof Goldberg (1995), this semantics will not be associated with arbitrarily largesyntactic objects, but constructed systematically on the basis of primitive recur-sive syntactic relationships. I will argue that once the most atomic predicationalrelations among basic formatives are taken into account, it is possible to seecomplex event-structural and argumental relations as being decomposable intosimpler ones, whichmoreover correspond to the simplest primitives of syntacticcombination (here taken to be Merge and a distinction between speciers andcomplements). Thus, the decomposition of verbal meaning will lead to a pro-posal concerning the functional sequence of the lowest part of the clause, anduniversal combinatoric semantics that goes along with it. Isolating this system-atic semantic combinatoric component of the grammar is only possible once aprincipled line is drawn between it and the lexical-encyclopedic and real-worldknowledge that goes along with every actual verb in context.

The problem of what constitutes the lexical information determined by a verbcarries us rst into the domains of argument-structure specicity and exibility,and event-structure/aktionsart specicity and exibility.What will emerge fromthe initial empirical discussion and summary from the literature in chapter 2 isthat a particular set of featural or combinatoric primitives seem to be implicatedin the linguistic generalizations we nd. The challenge of expressing the lexicalinformation in both these domains is to express both the exibility and the lim-itations that exist, and the interplay between different elements of the structurein a systematic way. In chapter 3, I make a specic proposal concerning thenature of what I will call the rst, or event-building, phase of the syntax (therst phase) and the relation between it and the lexicon. The central feature ofrst-phase syntax4 is that it decomposes the information classically seen to

4 I use the term rst phase here to imply logical priority. The event-building portion ofa proposition is assumed here to be prior to case marking/checking, agreement, tenseand modication in general. I make no assumptions about what the second phase is orwhat it should look like. Moreover, if the piece of syntax I am investigating here doesnot actually turn out to be a phase in the sense of Chomsky (2001) and others, it willnot greatly affect the proposals here, since I use no arguments from phase theory tocircumscribe the scope of my concerns.


reside within lexical items into a set of distinct categories with specic syntacticand semantic modes of combination. Lexical items in English will be seen tobe featurally complex, with their argument-structure properties and exibilityderiving ultimately from the association rules that link the particular featurebundle to the syntactic combinatoric system. In chapter 4, I will use the systemto spell out the decomposition of basic verbs in English in their different uses,including a reconceptualization of the classic conation-type verbs of Hale andKeyser, and an analysis of the double object construction. In chapter 5, resultaugmentation is considered. Here I examine in some detail the range of resul-tative and path augmentations in English, including prepositional phrases withmotion verbs, and adjectival resultatives. The verbparticle construction is alsodiscussed here as one of the most abstract morphemes in English contributingto the rst phase. I compare the particle construction in Germanic with com-pletive complex predicates in South Asian languages and lexical prexes inSlavic, arguing that the same underlying rst-phase syntax is involved, but withdifferent morphological composition. In chapter 6, I tackle the process of rst-phase syntax augmentation in the form of causativization, using the productivemorphology of direct and indirect causation in Hindi/Urdu as a test case. Thenal chapter is the conclusion and summarizes the proposals made in the book,and includes some speculations about how the system argued for interacts withthe rest of the combinatoric systemwith its more extended functional sequence.

2 The empirical ground

2.1 Selection versus variability

Over the years, it has been acknowledged that in addition to syntactic categoryinformation, lexical entries need to contain information related to their selec-tional properties. The specication of syntactic complementation can accountfor the difference between transitive and intransitive verbs, for example (1),or for the difference between the verbs that take CP complements vs. IPcomplements on the other (2).(1) (a) John saw the lizard./*John saw.

(b) *John dined the tortellini./John dined.

(2) (a) John hoped that the rain would fall./*John hoped the rain to fall.(b) *John got that the rain would fall./John got the rain to fall.

Syntactic selectional informationhowever, is not always deemed tobe enoughsince there seem to be generalizations related to the type of semantic participantthat make a difference to the linguistic behaviour of different verb types. So,for example, transitive experiencer subject verbs behave differently from tran-sitive verbs with agentive subjects (3) (Grimshaw 1979, Pesetsky 1982), andintransitive verbs with patient arguments (unaccusatives) seem to behave differ-ently from intransitive verbs with agent arguments (unergatives) (4) (Perlmutter1978, Williams 1980, Marantz 1984, inter alia).(3) (a) John fears tigers.

(b) John kills tigers.

(4) (a) The vase fell.(b) John danced.

To the extent that these differences are grounded in genuine linguisticbehaviour, and not simply a difference in real-world understanding, it seems

18

2.1 Selection versus variability 19

as if they need to be represented in the lexical entry of the predicates involved.Thematic roles are one way of dealing with generalizations of this type (Gruber1965, Baker 1988, Grimshaw 1990). Once thematic role information or seman-tic selectional properties are enshrined in the linguistic system in the form oftheta-marking, it is tempting to try to reduce facts that could be accounted forby syntactic selection to this kind of semantic selection as well, so that onlythat type of information need be present in the lexical entry (Grimshaw 1979,Pesetsky 1982, and recently in the context of distributedmorphologyHarley andNoyer 2000). However, there are two basic problems with making the reductionin this direction.

Firstly, there are serious doubts concerning the denability and empiricaladequacy of thematic role classications. The ultimate success of a theoryof -role types depends on nding linguistically legitimate natural classes ofarguments which can be systematically identied and studied.AsDowty (1989)has argued, the -role labels as traditionally formulated donotgive rise to naturallinguistic classes in terms of their syntactic or semantic behaviour (see alsoCroft 1998). In particular, Dowty (1989) has shown that many of the linguisticgeneralizations traditionally stated in terms of particular thematic relations,on further analysis have turned out to rely on distinctions within a particularthematic class1 or ondifferent semantic primitives altogether.2 In addition, usingprinciples like the thematic hierarchy to regulate mapping to the syntax doesnot always give the correct empirical results (cf. dative alternation verbs, psychpredicates with either experiencer objects or experiencer subject, or spray-loadalternations).

More recent argument-role classications have zeroed in on the fact thatthe factors that seem to make a difference to linguistic behaviour are cor-related with event structure or aktionsart properties. Vendlers 1967 articlepresenting theAristotelian classication of event types and relating it to classesof predicate in natural language is the source of much stimulating work onthe aspectual or event-structure classication of verbs (Dowty 1979, Taylor1977, Kenny 1963). While it is now understood that the original division

1 While the initial generalization was that do so substitution and the progressive bothpicked out the class of verbs with Agent subjects, it turned out that two differentnotions of Agent had to be distinguished: one characterized by the presence of motionor instigating change (for the do so test), and the other characterized by volition (forthe progressive) (Dowty 1989).

2 In the case of Dutch, auxiliary selection is argued in Zaenen (1993) to be sensitive not toThemesubject vs. Agent subject, but to the difference between denite and indenitechange of state (accomplishment vs. achievement).


into states, activities, achievements and accomplishments cannot corresponddirectly to what is specied in the lexicon, many theories attempt to use lower-level aspectual features that are derived from these larger natural classes.In particular, notions such as telicity/boundedness, dynamicity or durativ-ity have played an important role in subsequent theories of event-structuredecomposition and lexical classication. In general, many researchers haveattempted to classify verbs by means of their inherent aspectual properties(Grimshaw1990,Hoekstra 1984, 1992,Hoekstra andMulder 1990,Tenny1987,Levin and Rappaport Hovav 1995) as a way of capturing important linguisticgeneralizations.

Recently,many linguists have attempted an explicit aspectual classication ofthematic roles and relations themselves, often primarily to account for aspectualcompositional effects. Most notable in this class are the proposals introducinglower-level features such as +ADD-TO (which represents the verbs incre-mental or additive properties), +SQA (which encodes whether a specicquantity of matter is denoted by the DP) (Verkuyl 1989, 1993) or QUA (ageneral quantization property for both objects and events) and Mapping-to-Objects (a particular kind of thematic relation between verb and object) (Krifka1987) which can combine with the features of the lexical predicate to givetelicity under certain conditions. The aspectual thematic role in this sense isdened by the entailments about aspectual structure that it gives rise to (seealso Ramchand 1993, 1997). These classications are more successful than theclassical thematic role labels because they are denable on the basis of gen-uine linguistic diagnostics and are better at accounting for data such as thespray-load alternation and the unaccusative/unergative divide. Researchers inthe more traditional thematic role tradition have also increasingly used moreabstract and event-structure-based labels to categorize participant relations (cf.Baker 1997).

However, even these more satisfactory classications of participant relationshave to deal with the second problem facing any attempt to reduce lexicalclassication to semantic selection. This is the fact that argument-structureinformation is actually not nearly as rigid as lexical classication in generalwould imply. Any system of lexical classication of role types (whether clas-sically thematic or aspectual) has to face the reality of argument-structurevariability, in a fairly systematic and predictable form. For example, in English,a large class of verbs systematically occurs in an intransitive version with a sin-gle internal or theme-like argument, as well as a transitive version with bothan agent and a theme (the ergativeclass of verbs, according to the terminologyof Hale and Keyser 1987).


(5) (a) The glass broke.(b) John broke the glass.

These argument-structure alternations, whether mediated by morphologicalafxation or not, in English and other languages (e.g.middle formation, passive,causativization, etc.), seem to offer evidence for systematic lexicon-internalprocesses as an alternative to stipulated ambiguity with multiple lexical items(cf. Levin and Rappaport Hovav 1995). Unless such lexical redundancy rulesare postulated, representations of lexical information run the risk of failing tocapture pervasive generalizations concerning related/phonologically identicallexical items.

At the extreme end of the spectrum, the variability of behaviour seems so ram-pant as to be virtually unconstrained except by real-world knowledge. Considerfor example, the transitive creation/consumption verb such as eat in English,which can appear in the following grammatical environments with differentaspectual effects (cf. Folli and Harley 2004).

(6) (a) John ate the apple.(b) John ate at the apple.(c) The sea ate into the coastline.(d) John ate me out of house and home.(e) John ate.(f) John ate his way into history.

Data like these tempt one into the radical constructionalist approach of Borer(1998, 2005) or Marantz (1997b), whereby no lexical information is presentat all, but lexical items are inserted into syntactic contexts according to com-patibility with encyclopedic and real-world knowledge. Under this view, thegeneralizations reside in the systematic ways in which syntactic structuresare interpreted by the linguistic computational system, not in the informationspecied by lexical entries.

At the same time, however, verbal exibility is not completely general, as thedata in (7) and (8) show, otherwise the radical constructionalist view would beunavoidable.

(7) (a) John arrived.(b) *Bill arrived John.

(8) (a) Mary weighs 100 pounds.(b) *Mary weighs.


Flexibility exists on the level of aspectual specication as well, giving rise toproposals for lexicon-internal processes such as Levin and Rappaports (1998)template augmentation or event-type shifting (van Hout 2000, 2001). The twocore cases of event type-shifting involve (i) the adding of a causative subeventto an already possible event structure (as in (9)), or (ii) the adding of a telos toa process verb (as in (10)):3

(9) John jumped the horse over the fence.(10) John ate the porridge up.

But once again, these processes are not completely general since some verbsseem to resist causativization (11a), and others resist telic augmentation (11b):(11) (a) *John slept the baby.

(b) *John watched Mary bored/to boredom.Thus exibility in event structure and argument structure goes hand in hand

with more intangible limits and constraints.The strategy I will pursue is rst of all to reject the existence of formal

semantic selectional features in the lexicon, but attempt to account for whatrigidity there is in terms of purely syntactic or categorial features, made possi-ble by amore articulated view of the functional sequencewithin the verb phrase.I will show that once the selectional generalizations are properly understoodand isolated from the more heterogeneous and unsystematic felicity condi-tions based on encyclopedic meaning, they will be seen to be amenable torepresentation in terms of an articulated syntax with a systematic semanticinterpretation. This will allow a radical simplication of the architecture of thegrammar by reducing the set of combinatorial primitives and will account forimportant crosslinguistic data concerning the nature and exibility of lexicalitems.

The rst step is to establish and motivate the primitives that are empiricallynecessary in a decomposition of verbal meaning this is what the remainderof this chapter sets out to do. This sketch is intended as a basic outline of theimportant distinctions that need to be made in the face of the broadest empiricalpatterns, not as a complete exegesis of verb types. In chapter 4, after the the-oretical machinery has been introduced, I will return to the data in an attemptto offer diagnostics and to be more explicit about the syntax and semantics ofindividual verb types.

3 Here I am assuming that a telos can be added by a PP, adjectival resultatives and particles.


2.1.1 CausationThe approach Iwill take here is to argue that establishing the primitive role typesgoes hand in hand with establishing the primitive elements of event decomposi-tion, since participants in the eventwill only be denable via the role they play inthe event or subevent. The rst component of verbal meaning that has receivedmuch empirical support in the literature is that of causation. Causation hasbeen shown to be a relevant parameter in verbal differences and shows up veryoften as overt morphology within the verbal inventory of human languages (cf.Baker 1988, Hale and Keyser 1993, Ritter and Rosen 1998, Rappaport-Hovavand Levin 2000). Moreover, as I will argue next, it is implicated in the externalvs. internal argument distinction that has been used as a dening property ofverb classes within languages.

Ever since the unaccusative hypothesis of Perlmutter (1978), the existenceof an external argument or agent has been cited as criterial of a major divi-sion in (intransitive) verb types (e.g. Williams 1980, Marantz 1984). However,Rappaport-Hovav and Levin (2000) show convincingly that it is not agencyper se that determines class membership as either unaccusative or unergative.The following intransitives cited by them pass the diagnostics for unergativityin Italian, Dutch and Basque even though they do not possess arguments thatbring anything about by agentive action.

(12) glow, stink, spewEven in English, the fact that these verbs possess an external argument can be

demonstrated by their ability to takeXswayobjects under certain circumstances(examples 13) and also show an inability to causativize (examples 14).4

(13) (a) He stank his smelly way home.(b) The water spewed its way along the corridor.(c) John ran his way into history.

(14) (a) *Michael glowed Karenas face.(b) *We spewed the water out of the sink.(c) *We stank the dog by throwing him in the cesspit.(d) *John ran Mary by scaring her with a live mouse.

While it is true that many types of external argument can be distinguishedaccording to different semantic properties such as volitionality/agency (Butt1995 for Hindi/Urdu) or active vs. inactive causing (as in Doron 2003), they

4 These examples are taken from Rappaport-Hovav and Levin (2000).


all seem to be subclasses of argument that behave the same way with respectto our linguistic diagnostics for unaccusativity as shown above, and differentlyfrom internal arguments. Thus, I will accept the general intuition that there isan important primitive underlying the distinction between internal and exter-nal arguments (cf. Marantz 1984), but I will assume (with Rappaport-Hovavand Levin 2000 and many others) that the relevant abstract category is that ofinitiator. An initiator is an entity whose properties/behaviour are responsi-ble for the eventuality coming into existence. Thus, stinking has an externalargument which is the initiator by virtue of inherent properties of dirtiness orsmelliness; the water is the initiator of a spewing event by virtue of the factthat it has the requisite properties of kinetic energy; volitional agents haveintentions and desires that lead them to initiate dynamic events; instrumentalsubjects are entities whose facilitating properties are presented as initiating theevent because they allow it to happen. There is a sense in which all of thesethematic roles are just real-world instantiations of the more abstract conceptof causation.5

Among transitive verbs as well, external arguments can be volitional agents(15a, b), instrumentals (15c), abstract causes/sources (15df), showing thegenerality and abstractness of the external argument relation.

(15) (a) John broke the window.(b) John built that house.(c) The hammer broke the window.(d) The videotape from the secret camera demonstrated the truth of

the matter.(e) The storm broke the window.(f) Johns money built that house.

Im going to assume, therefore, that even though agency might be relevantfor felicity in certain circumstances, it does not directly determine syntacti-cally relevant class membership. The relevant notion here is that of causationor initiation, or more abstractly, the existence of a causing subevent, whichhas a DP role associated with it via the syntax (similar to Kratzer 1996) andwhich is specied more particularly by the lexical encyclopedic knowledge of

5 It is important to be clear that these are not claims about the real world, but abouthow human beings systematically interpret the situations they perceive in the world.Causation appears to be a very basic organizational category in these interpretationsand consistent with a number of different real-world possibilities.


the verb itself.6 I also leave it open exactly how the truth conditions of causa-tion/initiation should be specied. All that is necessary for our purposes is toestablish the existence of a primitive notion at this level of abstraction that cor-responds to the linguistic reality of how speakers conceive of events and theircomponents. The details of this position will be taken up again in chapter 3.

2.1.2 TelicityTelos or resultativity is also a component that has been shown to be isolable as aparameter in verbal meanings, and which has associated morphology and case-marking reexes in various languages (see, for example, Tenny 1987, Kiparsky1998, van Hout 1996, Ritter and Rosen 1998, Borer 1998). Semantically, ithas been widely argued that the combination of process and result createscomplex accomplishments (Parsons 1990, Pustejovsky 1991, Higginbotham2001). These two subevental components can be found separately or combinedwithin different verbal meanings, and can even be exploited to create morecomplex types out of simpler ones in many systems, cf. template augmentation(Levin and Rappaport 1998) or event type-shifting (van Hout 2000a, 2000b).

First, I wish to show that while there denitely are privileged relationshipsbetween certain arguments and certain aspectual subevents, the relationship isnot as straightforward as it might seem from only examining a subset of verbaltypes. In particular, there is no general one-to-one correspondence betweeninternal arguments and the semantic feature [+telic], even when the internalargument in question is quantized (in the sense of Krifka 1987, 1992). This iscontra the position taken in Kratzer (2004), Borer (2005) and van Hout (2000a).Specically, I will argue that there are two distinct kinds of aspectually sensitiveinternal arguments, and that quantization is only relevant for a subtype of oneof these.

The arguments for the lack of a simple relationship between the feature[+telic] and the internal argument go in both directions. First of all, the existenceof telicity does not actually imply the existence of a quantized internal argument(16b), or even an internal argument at all (16a).(16) (a) John stood up in a second. (no internal argument)

(b) They found gold in three hours. (mass term internal argument)

6 In the implementation that follows in chapter 3, I will not use the device of avours oflittle v (as in Harley 1995) to capture the different types of initiator found in language,but relegate such differences to the encyclopedic content of the root, or whatever lexicalelement lls that position.


Conversely, equally basic English examples can be used to show that theexistence of an internal argument does not imply telicity (not even when it isquantized) (17).(17) John pushed the cart for hours.

Kratzer (2004) builds on work by Kiparsky (1998) and Ramchand (1997)to offer a syntactic analysis that respects the semantic/aspectual correlates ofdifferential object case marking in Finnish and Scottish Gaelic respectively, andfound in many other languages. Her account makes a distinction between telic-ity and quantization, and conditions of culmination. In her account, objects aredirectly or indirectly responsible for establishing measures over the event, andneed to move to check their accusative feature (there seen as the uninterpretablecounterpart of a [+telic] feature) in a higher aspectual projection just outside v.In this sense, the account is fairly similar to that found in Borer (2005), wherethe quantized object must move to check its quantity feature against the quan-tity feature in the aspectual head dominating the verb phrase.7 Both accountsmust make extra stipulations to account for the cases where quantized objectsdo not in fact induce telicity (quantizedness on the part of the event), or caseswhere a nonquantized argument nevertheless occurs with a telic event. In thecase of Kratzer, this comes down to invoking covert measure phrases whichmust co-occur with objects that are not themselves measures; for Borer, inde-pendent (non-object-related) ways are found to check the quantization featureon the aspectual head. However, both of these strategies weaken the systemconsiderably, or rather, weaken the support for a syntactic featural connectionbetween quantization or accusative on the direct object and telicity or quantiza-tion on the verbal projection. The exceptions to the correlation, in my opinion,are central and normal enough that they cannot really be seen as exceptions.Instead, I propose to make some ner-grained distinctions in terms of how thedirect object maps onto the event, although I will preserve the intuition thatsome kind of event-topological mapping is criterial of direct objecthood.

The idea which I see as central to the distinctions we need to make is thatof a path to the event. By this I mean that dynamic verbs have a partwholestructure, as dened by our human perception of the notion of change. In thissense, dynamic events are generalized change analogues of spatial paths. Aswe saw in the previous section, a certain class of arguments of a dynamic

7 Kratzer rejects Krifkas (1987) notion of quantization as being exactly the right notionhere. So, in fact, does Borer although she retains the term, she offers a differentdenition of quantization than the one in Krifka (1987).


predicate can be distinguished as external they are related to the event asa whole, with a kind of abstract causational or initiational semantics. Internalarguments on the other hand are internal to the path structure of the event.However, I would argue there are a number of semantically distinct ways inwhich they can be so. The rst obvious case to consider is the argument thatis interpreted as undergoing the change asserted by the dynamic verb (cf. ageneral Undergoer role, after Van Valin 1990).

There are two distinct points to be made here. The rst is that even if wecharacterize an internal argument as one that crucially undergoes change, empir-ically it does not seem true that the changemust necessarily entail the attainmentof a nal state.

(18) widen, harden, melt, dry

The verbs shown above satisfy tests for unaccusativity in languages thatshow these clearly, and yet they are not obligatorily telic. A gap can widen butit doesnt necessarily become wide; the chocolate can melt, but it does not haveto become completely liquid.

(19) (a) The gap widened for three minutes (but still remained too narrowfor us to pass through).

(b) The chocolate melted for three minutes in the back seat of the car(before we rescued it).

While the attainment of a result state can give rise to telicity, mere gradualchange on the part of an argument is a distinct aspectual property and one whichis logically separable from the attainment of a result (although sometimes onecan be implied by context if the semantics of the verb is suitable), and hence iscompatible with a lack of temporal bound (see Hay, Kennedy and Levin 1999for an important discussion of the semantics of scales with regard to change-of-state verbs). Verbs which have an argument that undergoes a gradual change(without attainment of a denite result) often display unaccusative behaviourin the languages where the diagnostics are clear, indicating that they actuallyhave internal arguments in the relevant sense. Correspondingly in English, asRappaport-Hovav and Levin (2000) note, these verbs do not occur in the Xsway construction; and many of them do causativize.

(20) (a) John widened the gap between himself and his opponents.(b) Karena melted the chocolate in the pan.


It seems that what is crucial here is the notion of the argument undergo-ing some sort of identiable change/transition, for example whether it is withrespect to its location (21a), its state (21b), or its ullage8 (21c).(21) (a) The ball rolled down the hill.

(b) The mangoes ripened in the sun.(c) The bucket lled with rain.

In the case of transitive verbs, we nd direct objects that full this conditionof undergoing change as well: DPs can make good objects regardless ofwhether the change is that of location (22a), state (22b) or material properties(22c) (see Ramchand 1997 and Hay, Kennedy and Levin 1999).

(22) (a) John pushed the cart.(b) Mary dried the cocoa beans.(c) Michael stretched the rubber band.

The broad notion of undergoer (after Van Valin 1990) seems to be the oneresponsible for class membership here, and includes objects of verbs of changeof state like dry, as well as objects of verbs of translational motion like pushand drive. In other words, the existence of an undergoer does not necessarilyimply telicity, even when it is quantized (however we choose to dene that).

(23) (a) The document yellowed in the library for centuries.(b) John pushed the cart for an hour.(c) Mary dried the cocoa beans in the sun for an hour.

These objects are in a very general sense distinct from the causers/initiatorsof the previous section.What they all have in common is that they are undergo-ers of transitional states; this fact holds regardless of the internal denotationalconstitution of the DP in question.

What then of the notion quantized-ness or specied quantity that hasplayed such an important role so far in the literature on aspectual composition?Is there a class of verbs or class of objects that needs to be distinguished fromgeneral undergoers on the basis of their linguistic behaviour? Starting withVerkuyl (1972), the literature on aspectual composition has concentrated on aclass of creation/consumption verbs where the denotation of the DP object hasa direct effect on the aspectual nature of the verb phrase as a whole. So, to

8 A real, but underused, word of English referring to the volume by which a container isnot full.


recapitulate the data, in (24a), we see that a DP with homogenous referencesuch as a mass noun or a bare plural gives rise to a verb phrase without denitetemporal boundary,while aDPwith bounded or nonhomogenous reference suchas a singular or count term can give rise to a temporally bounded interpretationwith the very same verb (24b).(24) (a) Michael ate apples/ice cream for an hour/??in an hour.

(b) Michael ate the apple/ve apples for an hour/in an hour.These facts offer a tantalizing analogy between the denotational properties ofthe object and the denotational properties of the event that it gives rise to.However, mere transference of a feature of boundedness from object to event(as in, for example, Borer 2005) is a stipulation that does not rest on the semanticcompositional analysis of the phenomenon, and extends beyond the domain thatthe semantic compositional analysis is equipped to cover. We need to ask whysuch features can transfer not just syntactically (which we know to be possiblethrough general agreementprocesses), but in a semantically interpretable wayfrom one domain to another. In fact, Krifka (1992) offers just such an account:for a certain class of verbs the relation R between the verb and the objectsatises two crucial properties relating denotation of object and event,Mapping-to-Objects, and Mapping-to-Events. Given the satisfaction by R of these twoproperties, it can be shown that the right aspectual entailments follow. Whatis less often built into the systems implemented in the syntax is that, as hehimself acknowledges, the aspectual entailments follow only for the class ofverbs whose R relation has these particular properties specically these arejust the creation/consumption class of verbs.

Returning to our verbs describing change, it is only if the nature of thechange relates directly to the material extent of the object that the direct map-ping between object denotation and event denotation can be found. To transferboundedness from object to event in the general case of an undergoer is boththeoretically unfounded and empirically incorrect. In other words, if the transi-tions are related to the objects material extent, then quantizedness will producea telic entailment as in (24).

In fact, the creation/consumption type of transitive verb object ismore similarto the notion of path as found in examples with verbs of motion (25).(25) (a) John walked the trail.

(b) Mary ran along the beach.The notion of path or scale is now understood fairly well semantically andcross-cuts a number of distinct cognitive domains (see Schwarzschild 2002


on measures in general, Zwarts 2005 for spatial paths, Wechsler 2001 andKennedy 1999 for gradable states). As Hay, Kennedy and Levin (1999) pointout, the case of creation/consumption verbs is simply a special case of someattribute of the object contributing the measuring scale that is homomorphicwith the event. This property is shared by all paths, whether they are derivedfrom the object as in the case of creation/consumption, whether they come fromthe scale that can be inferred from a gradable adjective or whether it is a moreobvious physical path as contributed explicitly by a PP with a motion verb.Moreover, if one considers the motion verb push below, it is clear that path inthis sense is not a species of undergoer at all, but complementary to it: in (26),the path describes the ground that the undergoer traverses.

(26) John pushed the coconut along the beach.

Here the object DP, the coconut, is the undergoer because it is experiencingthe change of location, and the PP along the beach, is the path of motion.Logically, since the transitions are related to the objects change of location,then only the specication of a nal location will create telicity (27).

(27) John pushed the cart to the end of the garden.

If the transitions are related to the objects change of state, then only thespecication of the nal relevant statewill create telicity (28) (seeHay,KennedyandLevin 1999 for a detailed discussion of telicity effectswith this type of verb.)(28) Mary dried the cocoa beans bone dry in only twelve hours.

I would like to entertain the view that with the creation/consumption verbs,the DP argument does not itself travel some abstract path of change; it actuallydenes the path of change, and this is why it creates the quantization effects asnoted in the literature.

Thus, we really need to distinguish between undergoer and path if thediffering linguistic behaviour of these objects is to be understood.We also needto separate the predicational and relational properties described here from thepurely temporal notion of telicity.9 None of these verbs is obligatorily telic;

9 The relation between temporal bound and event-structure notions will be taken up inmore detail in the nal chapter. At that point, I will end up agreeing with Kratzer(2004) and Borer (2005) on the existence of an aspectual head related to actual temporalboundedness which sits outside the lowest (event-building) verbal domain. However,the notion of temporal bound will not be directly homomorphic with event-topologicalnotions as described here, and which form the basis of the core participant relationships.


they can be interpreted as telic as a result of entailments triggered by the natureof the direct object, and/or the specication of the nal state in the syntax (27and 26). I take the telicity effects in the class of creation/consumption verbswith quantized objects to be semantic entailments and not encoded in the lexicaldetermination of the verb or its syntactic reexes.

One other comment is in order concerning the nature of unboundedness. Wehave seen that with a creation/consumption verb, the homogenous nature of thedirect object translates into an unbounded or homogenous interpretation for theevent as a whole. However, this phenomenon, which is once again dependent onthe semantic properties of the verbal relation, is distinct from the more generalphenomenon of iterative readings, available for all verbs with plural objects.(29) (a) John ate TV dinners for years before learning to cook.

(b) Mary dried the dishes for hours before being released from duty.(c) Michael pushed the shopping carts to customers cars all day.(d) Peter threw away those empty jam jars for years before he realized

how useful they were.

This is, I believe, a completely independent phenomenon, as evidenced byits complete generality: the unboundedness emerges not because of the homo-geneity of the core event, but because the core event is being indenitelyrepeated/iterated once each for every individual within the plural set. As longas the actual cardinality of the plural object set is not determined by the con-text, such iteration will be unbounded. Notice that in the context of a denitenumber of objects (30a), or a plural object conceived of as a group (30b), theplural object can indeed be compatible with a PP requiring boundedness.(30) (a) John dried the dishes in an hour.

(b) Bill threw away the empty bottles in a ash.

So the effects here are not related specically to verb type, nor to generalquantizedness (just indenite plurality). Moreover, such effects are observedwith subjects as well as direct objects, with each individual in the group ofplural objects determining its own event of the relevant type.(31) (a) Tourists arrived at this pleasure spot for years.

(b) The buccaneers attacked this island for years.

This iteration of fully formed events is a case of external aspect, which needsto be excluded when analysing the phenomenon of aktionsart or event buildingthat will be the job of the lowest portion of the clause. These latter notions will


be taken up briey in the nal chapter, when the relation to external aspect andtense is discussed.

To summarize, then, we have isolated a class of verbs which represent aprocess or set of transitions, where one of the arguments (the undergoer) isthe subject of change.We also isolated a class of verbswhere the verbal changeis directly mapped on to the material extent of the object. I called these objectspaths, and in these cases, entailments concerning the events boundedness arisefrom the boundedness or unboundedess in the material extent of the object.

However, there are certain verbs that behave signicantly differently in beingobligatorily telic, even in English. They systematically reject the for an hourtest, in contrast to the verbs above where it is always possible to get an atelicreading.

(32) (a) John broke the stick in a second/*for seconds.(b) Mary arrived in two minutes/*for two minutes.(c) Michael found gold in just ten minutes.

Clearly, the telicity of this class of verbs needs to be represented differentlyfrom the telicity that sometimes arises from the semantic combination of theverb, its object (whether undergoer or path), and the presence of a nal state(implicit or explicit). The claim here is that these verbs resist the atelicity testbecause their objects are already dened as holders of a nal state. They dontjust undergo some change, they also end up in a nal state as specied by theverb itself. I will call this special type of role relation to the eventuality structureof the predicate the resultee.10 In the sentences in (32) above: the stick attainsa criterial identiable change of state so that its material integrity is ruptured;Mary attains a locational state as determined by the deictic context; the resultof Michaels actions must be that gold has been found. Notice that in (32c), theexistence of a result, and by extension telicity, is clear even thoughwe have useda mass term gold in object position. The result properties are thus properties ofthe verbal event structure, not of the interaction between direct objecthood andquantization.

Thus, in terms of subevental decomposition, we need to distinguish betweenprocess or change simpliciter, and the actual attainment of a result state or telos

10 Notice that resultees can also occur in unbounded events, if the unboundedness is cre-ated by externalmodication, as a part of external aspect. For example, as we saw above,a plural distributed object or subject can create an unbounded iteration of events. Thisphenomenon is independent of the core internal properties of the event as determininga nal or result state.


(as in much recent work, e.g. Pustejovsky 1991, Parsons 1990, Higginbotham2001). Correspondingly, the internal arguments that are the undergoers ofchange are distinguishable from the attainers of a nal state, although it is pos-sible (and indeed common) for a single argument to possess both properties.We have also seen the necessity of distinguishing paths from undergoers (orindeed resultees), because these former have special transfer properties con-cerning the homogeneity of object and event. During this discussion we havebeen careful also to distinguish the effects accruing because of these primi-tive event role relations from iteration of events and the general availability ofdistributive readings for all arguments.

2.1.3 Nonaspectual argumentsOne nal class of arguments needs to be considered now. So far we have lookedat participant roles that play a particular kind of relation specifying the subeven-tal decompositions of dynamic events: the initiator is the direct argumentrelated to the causing subevent (when it exists); the undergoer is the directargument related to the process subevent; and the resultee is the direct argu-ment related to the result state (when it exists)

Documents

[Gillian Catriona Ramchand] Verb Meaning and the L(BookZZ.org)